Hmmm .. I did this very successfully about 6 years ago - the important point for any dba is always have a backup strategy, I deployed 2 pci-e flash cards per node and split my tempdb files across the two cards.
I did extensive failure testing on what happened if a card failed and just as you would never deploy your sql server or databases to a non raid solution why would you even consider deploying tempdb to a single device?
I used this strategy on both failover clusters and Availability Groups, I never had an any issues and yes all my storage was monitored with suitable alerts.
I did write a white paper on this but i never published it generally as I assumed it wasn't really a big deal. It was for me a critical performance enhancement for some very busy 24 x 7 systems.
interesting that I don't ever recollect anyone writing about this previously. good article though.
[font="Comic Sans MS"]The GrumpyOldDBA[/font]